Revert workaround for createFirstGemmNegInfPadding on gfx11

[dhernandez0]

Motivation

This workaround was introduced for gfx11 here: #1699

This is likely a backend bug that has been fixed.

closes: https://github.com/ROCm/rocMLIR-internal/issues/1661

Technical Details

Remove workaround and its test.

Test Plan

e2e tests were added in #1699 that check if this works. Make sure the tests pass.

Test Result

Test pass.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull Request Overview

This PR reverts a workaround that was implemented for gfx11 architecture to address a backend bug that has since been fixed. The change removes forced unrolling behavior and associated test code that was added as a temporary measure.

Key changes:

• Removes the isGfx11 parameter and conditional logic from the createFirstGemmNegInfPadding function
• Restores forceUnroll=true behavior for all architectures
• Removes the test case specifically designed to validate the gfx11 workaround

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
mlir/test/Dialect/Rock/gridwise_attention_accel_lowering.mlir	Removes test case for gfx11 workaround validation
mlir/lib/Dialect/Rock/Transforms/GridwiseGemmToBlockwise.cpp	Removes gfx11-specific parameter and logic, restores original forceUnroll behavior

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[justinrosner]

Changes look good to me

[umangyadav]

@djramic could this be an issue with the upstream merge failing on navi3x ?